Key Takeaways
- Forget typing. Voice prompting AI at work is coming, and your CEO might want it. Rahul Sunwalker sees it as a new accountability layer: “I can audit what people are typing in the office” and “If they're like talking to the AI... it's more more accountability.”
- Your frantic "don't hallucinate" prompts are probably wasted. John Coogan says AI models have improved so much that "all that's been either baked into the pre-training or post-training or it's like even in the system prompt already."
- The real shift isn't just what you say, but how you say it. Coogan stopped "yelling at AI" by May 2026, realizing it's better to guide and reassure.
- This softer approach is formalized in “John Coogan's Cognitive Behavioral Therapy (CBT) Approach to AI Prompting,” a method designed to treat AI less like a servant and more like a collaborator.
The John Coogan's Cognitive Behavioral Therapy (CBT) Approach to AI Prompting
Type: method
Name: John Coogan's Cognitive Behavioral Therapy (CBT) Approach to AI Prompting
Components:
- Initial State (Pre-2025): Yelling at the AI, being demanding, focusing on “don't make mistakes, don't hallucinate.”
- Shift to CBT Approach (2025-2026): Completely giving up on yelling. Instead, reassuring the AI, setting it up for success by encouraging it, e.g., 'You can do this. Like think about think about you know you're you're like Palmer Lucky. And you can design this like new piece of hardware that's never existed before.'
- Post-Training/System Prompt Reliance: Recognizing that concerns about mistakes and hallucinations are increasingly 'baked into the pre-training or post-training or it's like even in the system prompt already,' reducing the need for explicit anti-hallucination prompts.
When This Works (and When It Doesn't)
This method aims to foster a more collaborative and effective interaction with AI, by treating it less as a tool to be commanded and more as an entity to be guided and encouraged, especially as models become more sophisticated at handling common pitfalls. Coogan's "CBT" approach shines when you're pushing AI for creative, open-ended tasks where generic, command-line prompts hit a wall. Think brainstorming novel product features or designing a market entry strategy for an unknown region. It works because it taps into the AI's vast knowledge in a less constrained way, encouraging it to "think" more broadly.
However, this method might be overkill for simple, deterministic tasks like data extraction or code formatting where precise, direct instructions are still king. You don't need to encourage an AI to write a regex. Similarly, if your AI model is older, less advanced, or hasn't received extensive post-training for hallucination control, you might still need to include explicit guardrails. Coogan's insight about baked-in safety applies more to the latest, most capable models. Don't reassure a basic chatbot; it might just misunderstand.
What to Do With This
Next time you're stuck on a complex problem and the AI's initial output feels generic, shift your prompting style. Instead of a terse demand like "Generate 5 marketing taglines for a new B2B SaaS product," try Coogan's CBT. Frame it like this: "Hey AI, you're the leading expert in SaaS growth. I know you can come up with something truly innovative here. Imagine you're advising a pre-seed startup, trying to cut through the noise. Give me 5 taglines that are sharp, memorable, and speak directly to a busy CTO's pain points. Really think outside the box." Then, watch how the AI responds. You might just unlock a level of creativity you didn't know it had.